Chemical Reaction Optimization on Minerva Olympus Suzuki Coupling i (complete matched replay suite)
88.5Final Best ScoreCARE
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| CARE2026.06 | 88.5 | 84.4 | 0.7 | 100 | |
| CAREAblation=w/o recovery proposal mode2026.06 | 87.6 | 82.8 | 1.6 | 93 | |
| CAREAblation=w/o intervention gate2026.06 | 87.3 | 82.7 | 1.9 | 90 | |
| CAREAblation=w/o Policy Planner2026.06 | 87.2 | 82 | 2 | 87 | |
| Public incumbent2026.06 | 80 | 73.4 | 9.2 | 47 | |
| CAREAblation=w/o frontier proposal mode2026.06 | 78.2 | 72.7 | 11.1 | 30 | |
| TPE-style BO2026.06 | 77.6 | 68.3 | 11.6 | 20 | |
| LMABO-style LLM-BO2026.06 | 76.6 | 66.5 | 12.6 | 17 | |
| GP-UCB2026.06 | 75.4 | 66.7 | 13.9 | 37 | |
| BayBE/BoFire-style mixed BO2026.06 | 75.3 | 67.7 | 14 | 0 | |
| Generic surrogate BO2026.06 | 74.9 | 63.7 | 14.4 | 23 | |
| Gryffin-style categorical BO2026.06 | 73.6 | 67 | 15.7 | 17 | |
| EDBO-style descriptor GP-EI2026.06 | 73.4 | 65.7 | 15.9 | 20 | |
| GP-EI2026.06 | 72.7 | 65.5 | 16.5 | 20 | |
| BoTorch-style GP-LogEI2026.06 | 72.5 | 65.4 | 16.8 | 20 | |
| Chemistry descriptor BO2026.06 | 69.9 | 65.3 | 19.3 | 3 | |
| Fixed public heuristic2026.06 | 69.1 | 66.5 | 20.1 | 13 | |
| Ungated LLM-evolving policy2026.06 | 68.8 | 61.4 | 20.4 | 13 | |
| Categorical empirical-Bayes UCB2026.06 | 63.9 | 59.2 | 25.4 | 3 | |
| SMAC-style RF-EI2026.06 | 63.7 | 56.6 | 25.6 | 0 | |
| Non-evolving LLM policy2026.06 | 63.3 | 58.8 | 25.9 | 3 | |
| Random2026.06 | 61.1 | 56.7 | 28.1 | 0 | |
| Stratified random2026.06 | 60.4 | 57.5 | 28.9 | 0 |